From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 53AE5A0093 for ; Fri, 17 Jun 2022 19:22:39 +0200 (CEST) Received: from [217.70.189.124] (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id 4A903410FC; Fri, 17 Jun 2022 19:22:39 +0200 (CEST) Received: from mga14.intel.com (mga14.intel.com [192.55.52.115]) by mails.dpdk.org (Postfix) with ESMTP id 8EB2940F19; Fri, 17 Jun 2022 19:22:36 +0200 (CEST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1655486557; x=1687022557; h=from:to:cc:subject:date:message-id:mime-version: content-transfer-encoding; bh=yl+KaPmnC4jUzPa1FLILgPu4A6GnrN0f93xMPZ7d7EQ=; b=H/0BSAgvaEhtPsMyXmOkHUY2TyWuSc6oG52dOkDQcugXjajvMRXnqkRu 8CtAYgjKTUmd/mnKrvmCwY/wQsXsCffpdFYLN1q3Jg51vDtFaonlWABVn BlqkZTVKwRT8nYdnyrKD7WIUco7WAInLF9eWxsCiP8DiVJYGEaWoHvIy5 AFbkPqigS281dZeKcXA17ebB1aU2HhtobHxQPo2yyMronChDHgYXfH6HQ Itg7udRURV072rf+0zpLtDzRsIFADH9POyhSqovEZ/Jw74AwOu+X29k4O WPbUy3VuQ1f4eKu/gsLZa8vax/EOZ+TIAZPz7n1EAxn/C61h90bqrTrPU g==; X-IronPort-AV: E=McAfee;i="6400,9594,10380"; a="279535292" X-IronPort-AV: E=Sophos;i="5.92,306,1650956400"; d="scan'208";a="279535292" Received: from fmsmga003.fm.intel.com ([10.253.24.29]) by fmsmga103.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 17 Jun 2022 03:59:24 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.92,306,1650956400"; d="scan'208";a="675451635" Received: from silpixa00401385.ir.intel.com (HELO silpixa00401385.ger.corp.intel.com.) ([10.237.223.181]) by FMSMGA003.fm.intel.com with ESMTP; 17 Jun 2022 03:59:23 -0700 From: Bruce Richardson To: dev@dpdk.org Cc: yongxin.liu@windriver.com, Bruce Richardson , stable@dpdk.org, Kevin Laatz , Conor Walsh , Chengwen Feng Subject: [PATCH] dma/idxd: fix AVX2 code in non-datapath functions Date: Fri, 17 Jun 2022 11:59:20 +0100 Message-Id: <20220617105920.354707-1-bruce.richardson@intel.com> X-Mailer: git-send-email 2.34.1 MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-BeenThere: stable@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: patches for DPDK stable branches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: stable-bounces@dpdk.org While all systems which will use the idxd driver for hardware will support AVX2, if the driver is present the initialization functions e.g. to register logs, will be called on all systems - irrespective of HW support. This can cause issues if the system running DPDK does not have AVX2, and the compiler has included AVX instructions in the initialization code. To fix this, remove AVX2 instruction set from the whole build of the driver. Instead, we add "target(avx2)" attribute to all datapath functions, so those - and only those functions - will having AVX2 instructions in them. Bugzilla ID: 1038 Fixes: 3d36a0a1c7de ("dma/idxd: add data path job submission") Cc: stable@dpdk.org Signed-off-by: Bruce Richardson --- drivers/dma/idxd/idxd_common.c | 23 +++++++++++++++++++++++ drivers/dma/idxd/meson.build | 1 - 2 files changed, 23 insertions(+), 1 deletion(-) diff --git a/drivers/dma/idxd/idxd_common.c b/drivers/dma/idxd/idxd_common.c index ea6413cc7a..c77200a457 100644 --- a/drivers/dma/idxd/idxd_common.c +++ b/drivers/dma/idxd/idxd_common.c @@ -13,12 +13,23 @@ #define IDXD_PMD_NAME_STR "dmadev_idxd" +/* systems with DSA all support AVX2 so allow our data-path functions to + * always use at least that instruction set + */ +#ifndef __AVX2__ +#define __use_avx2 __attribute__((target("avx2"))) +#else +#define __use_avx2 +#endif + +__use_avx2 static __rte_always_inline rte_iova_t __desc_idx_to_iova(struct idxd_dmadev *idxd, uint16_t n) { return idxd->desc_iova + (n * sizeof(struct idxd_hw_desc)); } +__use_avx2 static __rte_always_inline void __idxd_movdir64b(volatile void *dst, const struct idxd_hw_desc *src) { @@ -28,6 +39,7 @@ __idxd_movdir64b(volatile void *dst, const struct idxd_hw_desc *src) : "memory"); } +__use_avx2 static __rte_always_inline void __submit(struct idxd_dmadev *idxd) { @@ -74,6 +86,7 @@ __submit(struct idxd_dmadev *idxd) _mm256_setzero_si256()); } +__use_avx2 static __rte_always_inline int __idxd_write_desc(struct idxd_dmadev *idxd, const uint32_t op_flags, @@ -112,6 +125,7 @@ __idxd_write_desc(struct idxd_dmadev *idxd, return job_id; } +__use_avx2 int idxd_enqueue_copy(void *dev_private, uint16_t qid __rte_unused, rte_iova_t src, rte_iova_t dst, unsigned int length, uint64_t flags) @@ -126,6 +140,7 @@ idxd_enqueue_copy(void *dev_private, uint16_t qid __rte_unused, rte_iova_t src, flags); } +__use_avx2 int idxd_enqueue_fill(void *dev_private, uint16_t qid __rte_unused, uint64_t pattern, rte_iova_t dst, unsigned int length, uint64_t flags) @@ -136,6 +151,7 @@ idxd_enqueue_fill(void *dev_private, uint16_t qid __rte_unused, uint64_t pattern flags); } +__use_avx2 int idxd_submit(void *dev_private, uint16_t qid __rte_unused) { @@ -143,6 +159,7 @@ idxd_submit(void *dev_private, uint16_t qid __rte_unused) return 0; } +__use_avx2 static enum rte_dma_status_code get_comp_status(struct idxd_completion *c) { @@ -163,6 +180,7 @@ get_comp_status(struct idxd_completion *c) } } +__use_avx2 int idxd_vchan_status(const struct rte_dma_dev *dev, uint16_t vchan __rte_unused, enum rte_dma_vchan_status *status) @@ -180,6 +198,7 @@ idxd_vchan_status(const struct rte_dma_dev *dev, uint16_t vchan __rte_unused, return 0; } +__use_avx2 static __rte_always_inline int batch_ok(struct idxd_dmadev *idxd, uint16_t max_ops, enum rte_dma_status_code *status) { @@ -224,6 +243,7 @@ batch_ok(struct idxd_dmadev *idxd, uint16_t max_ops, enum rte_dma_status_code *s return -1; /* error case */ } +__use_avx2 static inline uint16_t batch_completed(struct idxd_dmadev *idxd, uint16_t max_ops, bool *has_error) { @@ -275,6 +295,7 @@ batch_completed(struct idxd_dmadev *idxd, uint16_t max_ops, bool *has_error) return ret; } +__use_avx2 static uint16_t batch_completed_status(struct idxd_dmadev *idxd, uint16_t max_ops, enum rte_dma_status_code *status) { @@ -366,6 +387,7 @@ batch_completed_status(struct idxd_dmadev *idxd, uint16_t max_ops, enum rte_dma_ return ret; } +__use_avx2 uint16_t idxd_completed(void *dev_private, uint16_t qid __rte_unused, uint16_t max_ops, uint16_t *last_idx, bool *has_error) @@ -383,6 +405,7 @@ idxd_completed(void *dev_private, uint16_t qid __rte_unused, uint16_t max_ops, return ret; } +__use_avx2 uint16_t idxd_completed_status(void *dev_private, uint16_t qid __rte_unused, uint16_t max_ops, uint16_t *last_idx, enum rte_dma_status_code *status) diff --git a/drivers/dma/idxd/meson.build b/drivers/dma/idxd/meson.build index f1396be945..dcc0a297d7 100644 --- a/drivers/dma/idxd/meson.build +++ b/drivers/dma/idxd/meson.build @@ -5,7 +5,6 @@ build = dpdk_conf.has('RTE_ARCH_X86') reason = 'only supported on x86' deps += ['bus_pci'] -cflags += '-mavx2' # all platforms with idxd HW support AVX sources = files( 'idxd_common.c', 'idxd_pci.c', -- 2.34.1